PuSH - Publikationsserver des Helmholtz Zentrums München

Export: TXT Text RIS Endnote (RIS) BIB BibTeX
Tran, M. ; Lahiani, A.* ; Dicente Cid, Y.* ; Boxberg, M.* ; Lienemann, P. ; Matek, C. ; Wagner, S. ; Theis, F.J. ; Klaiman, E.* ; Peng, T.

B-Cos Aligned Transformers Learn Human-Interpretable Features.

In: (26th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), Vancouver, CANADA, 8-12 October 2023). Berlin [u.a.]: Springer, 2023. 514-524 (Lect. Notes Comput. Sc. ; 14227 LNCS)
Postprint DOI
Open Access Green
Vision Transformers (ViTs) and Swin Transformers (Swin) are currently state-of-the-art in computational pathology. However, domain experts are still reluctant to use these models due to their lack of interpretability. This is not surprising, as critical decisions need to be transparent and understandable. The most common approach to understanding transformers is to visualize their attention. However, attention maps of ViTs are often fragmented, leading to unsatisfactory explanations. Here, we introduce a novel architecture called the B-cos Vision Transformer (BvT) that is designed to be more interpretable. It replaces all linear transformations with the B-cos transform to promote weight-input alignment. In a blinded study, medical experts clearly ranked BvTs above ViTs, suggesting that our network is better at capturing biomedically relevant structures. This is also true for the B-cos Swin Transformer (Bwin). Compared to the Swin Transformer, it even improves the F1-score by up to 4.7% on two public datasets.
Altmetric
Tags
Anmerkungen
Besondere Publikation
Auf Hompepage verbergern

Zusatzinfos bearbeiten
Eigene Tags bearbeiten
Privat
Eigene Anmerkung bearbeiten
Privat
Auf Publikationslisten für
Homepage nicht anzeigen
Als besondere Publikation
markieren
Publikationstyp Artikel: Konferenzbeitrag
Schlagwörter Explainability ; Interpretability ; Self-attention ; Transformer
Sprache englisch
Veröffentlichungsjahr 2023
HGF-Berichtsjahr 2023
ISSN (print) / ISBN 0302-9743
e-ISSN 1611-3349
Konferenztitel 26th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI)
Konferzenzdatum Vancouver, CANADA
Konferenzort 8-12 October 2023
Zeitschrift Lecture Notes in Computer Science
Quellenangaben Band: 14227 LNCS, Heft: , Seiten: 514-524 Artikelnummer: , Supplement: ,
Verlag Springer
Verlagsort Berlin [u.a.]
Institut(e) Helmholtz Artifical Intelligence Cooperation Unit (HAICU)
Institute of AI for Health (AIH)
Institute of Computational Biology (ICB)
POF Topic(s) 30205 - Bioengineering and Digital Health
Forschungsfeld(er) Enabling and Novel Technologies
PSP-Element(e) G-530006-001
G-540007-001
G-503800-001
Förderungen Helmholtz Association under the joint research school "Munich School for Data Science"
DOI 10.1007/978-3-031-43993-3_50
WOS ID 001109637500050
Scopus ID 85174717745
Erfassungsdatum 2023-11-28
[➜Korrektur melden]